Everything P-Video-2 can do

P-Video-2 is Pruna AI's video model. It makes the sound with the picture, from the same prompt: voices whose lips match the lines you quote, effects, music. It makes clips of 1 to 20 seconds at 720p or 1080p, can start from a first frame, end on a last one, or follow a soundtrack of your own, and has a cheaper draft mode for trying ideas.

Every clip below was made by the step under it. Change a placeholder and press Gen to make your own.

Quote the line

Put each spoken line in quotes and the clip says it word for word, the mouth in time. For two people, tie each line to a place in the frame and something they wear, and say they take turns. Write what happens after the line too, so the clip doesn't end on a frozen face.

Name the delivery as you would name a light: dry deadpan, three short phrases, over a beat. A line someone could say in one breath comes out cleanest; a long one gets rushed.

Say what it sounds like

The sound comes on by default, and if the prompt says nothing about it the model picks something. Write an Audio: line, and say how loud: a sound asked for quietly tends to be dropped. The same forge, three sound lines:

Cuts on cue

One clip can hold several shots. Number them ("First shot… Cut to a second shot…") and give each its own framing, close or wide, so the cut reads as an edit. When you want one take, say "one continuous shot, no cuts".

Start from a still

Give it a first frame and the clip starts there: the face, the room and the colors stay as they are, and the prompt only says what moves. The still decides the clip's shape. This one was made with FLUX 3 Image; drop in your own photo instead.

Both ends

Give it a first and a last frame and it finds the path between them. Change one thing between the two, here the time of day, and make the last frame from the first, so everything else matches. Two frames with nothing physical between them come out as a dissolve.

With both ends fixed, Duration sets the pace: a few seconds plays as a transition, ten or more as a time-lapse.

Use the same picture as both frames and the clip ends exactly where it began: a loop with no seam. Ask for motion that settles, like steam or rain, not a walk. With Audio off, the loop is silent, which suits a page that autoplays muted.

Draft first

Draft makes the same clip at lower quality for about 40% less. Use it to check the idea: the look, the light, the move. Turning it off makes a new take, not a sharper copy of the draft, so judge the fine detail and any speaking at full quality. 1080p costs twice 720p.

Made for the feed

For a vertical clip, frame the subject low and leave the top for the caption. One camera move per clip holds; several in a few seconds bend the room. Movement inside a still frame is the safest life you can add.

To build on any step, press Mix to copy its recipe into your own page.